Page 183 - Kaleidoscope Academic Conference Proceedings 2020
P. 183
AN AI-BASED OPTIMIZATION OF HANDOVER STRATEGY IN NON-TERRESTRIAL
NETWORKS
Chenchen Zhang; Nan Zhang; Wei Cao; Kaibo Tian; Zhen Yang
State Key Laboratory of Mobile Network and Mobile Multimedia Technology, ZTE Corporation, China
ABSTRACT Previous studies generally make handover decisions based on
one or more predefined criteria. The most commonly used
A complicated radio resource management, e.g., handover criteria include elevation angle [2], remaining service time[3]
condition, will be suffered by the user in non-terrestrial and number of free channels [4], which correspond to signal
networks due to the impact of high mobility and hierarchical strength, handover number and satellite burden, respectively.
layouts which co-exist with terrestrial networks or various But these methods cannot get an overall optimization. In
platforms at different altitudes. It is necessary to optimize [5], an overall optimization method is proposed by modelling
the handover strategy to reduce the signaling overhead the handover process by a directed graph. Each satellite is
and improve the service continuity. In this paper, a new denoted by a node, then the best handover strategy is obtained
handover strategy is proposed based on the convolutional by searching the shortest path. However, in [5] each satellite
neural network. Firstly, the handover process is modeled node is invariable during the handover process. A UE needs to
as a directed graph. Suppose a user knows its future signal perform handover as soon as entering the coverage of another
strength, then it can search for the best handover strategy beam and cannot choose an appropriate time. Besides, the
based on the graph. Secondly, a convolutional neural UE needs to predict its coverage condition in a future time to
network is used to extract the underlying regularity of the construct the graph, which may bring unexpected error and
best handover strategies of different users, based on which is beyond the capability of a standard 5G UE.
any user can make near-optimal handover decision according
to its historical signal strength. Numerical simulation shows In recent years, some artificial intelligence (AI) techniques
that the proposed handover strategy can efficiently reduce the have been applied to search overall optimization on handover.
handover number while ensuring the signal strength. The most often used technique is the Q-learning [6], [7], [8],
Keywords - Convolutional neural network, directed graph, which is typical model-free reinforcement learning (RL). In
handover, low earth orbit, non-terrestrial network Q-learning, some properties of a UE are defined as its state,
and the handover operation is defined as action. Numerical
simulation is used to iteratively train the Q-table (the reward
1. INTRODUCTION
of each action for each state) until its convergence. Then the
The non-terrestrial network (NTN) has been regarded as a UE is able to decide whether to perform handover according
supplement to the fifth generation (5G) terrestrial mobile to its state. Furthermore, the Q-table can be replaced by a
network for providing global coverage and service continuity neural network for an infinite number of states. In paper [8],
[1]. Compared with terrestrial networks, the handover in the handover in an LEO scenario is optimized by Q-learning.
NTN is more frequent and complex. In this paper, a handover The state of a UE is composed by its position, accessible
optimization method is proposed and applied to a typical satellites and whether handover is processed in this time
NTN scenario, i.e., low earth orbit (LEO) satellite network. slot. In each time slot, the UE is required to know its own
A LEO is an orbit around the earth with an altitude between state and will choose a satellite for handover, which is a
500 km and 2000 km [1]. Compared with geostationary earth really strong requirement for an ordinary UE. Besides RL, a
orbit satellites, the LEO satellites have much lower path-loss recursive neural network (RNN) also can be used for handover
and propagation delay. Therefore, the third generation optimization. Papers [9] and [10] apply RNN for handover
partnership project (3GPP) NTN study item has regarded optimization in terrestrial millimeter wave mobile systems
the LEO satellites as the key component to provide global and vehicular networks, respectively. However, in an LEO
broadband Internet access. Suppose the orbit is circular, the scenario the beam switch is fast, and the signal series of one
satellite will move around the earth in a constant velocity beam may be too short for the RNN to make decisions.
which is inversely proportional to the square root of the orbit
altitude. Because of the low altitude, the LEO satellites have In practical terms, a handover strategy with a low requirement
high speed with respect to the earth, and a terrestrial user for UE capability is desired to reduce the handover number
equipment (UE) needs to frequently switch to new beams while ensuring the reference signal received power (RSRP).
to keep connectivity. In order to ensure the quality of the In this paper, a convolutional neural network (CNN) based
Internet service, the optimization for handover strategy needs handover strategy optimization is proposed. Firstly, a number
to be carefully investigated. of UEs are randomly generated within the coverage of a
978-92-61-31391-3/CFP2068P @ ITU 2020 – 125 – Kaleidoscope